NVIDIA AI & NIMopen-weight
Released: 2026-06-04Artificial Analysis

Nemotron 3 Ultra

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...

Intelligence Index
11.8/ 100
Artificial Analysis Intelligence Index (independent composite)
Throughput Speed
203tok/sec
Output streaming throughput
Time to First Answer
4.9s
Time to first answer token, including any thinking
Cost to Run the Index
2.21USD
Artificial Analysis cost of the full Index run

Measured results · Artificial Analysis

Independent evaluations of Nemotron 3 Ultra (Reasoning configuration). Values are reported as published; nothing here is estimated or derived. Snapshot 2026-11-15.

Intelligence Index
11.8
Cost to run the Index
$2.21
Time per task
4.2 min
Omniscience index
3.4
Reasoning model
Yes
Cached-input price
$0.12 / 1M

Agentic work

Multimodal & science

  • AA-3D (Elo)583 Elo
  • Code2Video-AA (Elo)304 Elo
Full analysis on Artificial Analysis

Headline Evaluation Breakdown

Six independently measured evaluations (Artificial Analysis)
Reasoning · CritPt v2 (Hard)15.0%
Coding & agents · Terminal-Bench 4.00.5%
Long context · AA-LCR79.3%
Knowledge · Humanity's Last Exam28.4%
Instruction following · IFBench81.4%

Sourced Benchmark Evaluations (23)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
AA-LCR (Long Context Reasoning)Active79.3%
93
2026-11-15independentSource ↗
EnterpriseOps-Gym-AAActive28.9%
93
2026-11-15independentSource ↗
AA-AnalystAgentActive6.3%
93
2026-11-15independentSource ↗
AutomationBench-AAActive3%
93
2026-11-15independentSource ↗
MLCR-AA (Medical Long Context Reasoning)Active5.6%
93
2026-11-15independentSource ↗
GDP.pdfActive5%
93
2026-11-15independentSource ↗
FrontierCode-AA 2.0Active1.1%
93
2026-11-15independentSource ↗
AA-Omniscience (Non-Hallucination)Active77.4%
93
2026-11-15independentSource ↗
AA-Omniscience (Accuracy)Active21.2%
93
2026-11-15independentSource ↗
HLE-DiamondActive8.5%
93
2026-11-15independentSource ↗
CritPt v2 (Hard)Active15%
93
2026-11-15independentSource ↗
CritPtActive3.1%
93
2026-11-15independentSource ↗
FrontierCodeActive10.5%
98
2026-11-15independentSource ↗
Terminal-Bench-Science 0.1Active0%
93
2026-11-15independentSource ↗
Terminal-Bench 4.0Active0.5%
93
2026-11-15independentSource ↗
Terminal-Bench 2.1Active53.9%
93
2026-11-15independentSource ↗
Terminal-Bench HardActive36.4%
93
2026-11-15independentSource ↗
τ³-BankingActive14.2%
93
2026-11-15independentSource ↗
τ²-Bench TelecomActive83.3%
93
2026-11-15independentSource ↗
SciCodeActive40.3%
93
2026-11-15independentSource ↗
IFBenchActive81.4%
93
2026-11-15independentSource ↗
Humanity's Last ExamActive28.4%
93
2026-11-15independentSource ↗
GPQANearing Saturation86.7%
78
2026-11-15independentSource ↗
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run nvidia-nemotron-3-ultra

# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nvidia-nemotron-3-ultra",
    "messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
  }'

Evaluation & Sourcing Notes

Scores listed for Nemotron 3 Ultra represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.

All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.